Fix get_contact_count sometimes being incorrect in _process

[mihe]

Fixes #88018.

This replaces the usage of body_get_direct_state in RigidBody*D.get_contact_count with a cached value that's populated as part of RigidBody*D._sync_body_state, similar to other states in the physics bodies, which prevents _process from accessing a yet-to-be-synced future contact count.

This also fixes the inability to use get_contact_count from _process when using the physics/*d/run_on_separate_thread project setting.

[AThousandShips]

I'm assuming no operations like move_and_collide or anything else can change the collision count? And that it's fully safe to just cache it

[mihe]

Yes, that's correct. Contact counts are solely generated/modified during the simulation step.

[akien-mga]

Makes sense to me.

[akien-mga]

Thanks!